Enhancing Prediction of Prosthetic Fingers Movement Based on sEMG using Mixtures of Features and Random Forest

Abstract

As the recent development of the prosthetic control systems, it is necessary to develop the sub-components such as myoelectric control system. This sub-system is used to acquire electromyogram (EMG) signals from a person's muscles and convert it to movements to control prosthetic hands or fingers. Recently, researchers started focusing on providing feature extraction methods for both time domain and frequency domain for predicting either hand or finger movements. This paper proposes a new mixed feature set for both time and frequency domain for the classification of surface EMG (sEMG) into ten classes for controlling prosthetic finger movements. The features are based on enhanced statistical features extracted directly from a non-overlapped window with a predefined length which was selected carefully from the sEMG directly. Ten classes of individual and combined finger movements are to be recognized by using two sEMG channels from two electrodes are fixed on the human forearm to employ effective knowledge discovery and pattern recognition algorithms to enhance the recognition and classification accuracy. and the other features are statistical features extracted from Instantaneous frequency of the signal to utilize a suitable classifier helps detecting and recognizing the pattern from sEMG signals of different classes of the fingers movements either combined or single movement to enhance the classifier performances and to find the whole class of all extracted window ,a majority voting technique was applied.The method used the random forest as a classifier to build the classifier model which achieved an accuracy of 93.75% and sensitivity of 93.73% and specificity of 99.31%.